Luke Ditria's Raspberry Pi Bird-Spotting Camera Now Handles Long-Term Solar Operation

Maker Luke Ditria has once again revisited his Raspberry Pi Zero 2 W-powered bird-spotting wildlife camera project, this time upgrading it with a 3D-printed enclosure and a solar panel with custom-designed photovoltaic HAT add-on for full off-grid operation.

"I started this project last year and I've made a few videos about the previous versions, but I'm back and with some pretty big updates," Ditria says of the project's evolution. "The device is still built around Raspberry Pi's so-called AI Camera, which uses the Sony IMX500 intelligent image sensor. This sensor integrates a camera and a small neural network accelerator into a single chip, letting you run a small neural network model directly on the camera itself to process images at up to 30 frames a second. I've been using this sort of edge-compute AI detection for some time now, primarily for autonomously detecting and identifying wildlife."

The first version of Ditria's autonomous wildlife camera was unveiled a year ago, using a Raspberry Pi 5 single-board computer and the Raspberry Pi AI HAT+ Hailo-based accelerator board to run a custom-trained You Only Look Once (YOLO)-based neural network model designed to recognize around 30 different birds. The first major revision to the project shrunk both the power and space requirements considerably, swapping the Raspberry Pi 5 out for the smaller and cheaper Raspberry Pi Zero 2 W and shifting the neural model to the Raspberry Pi AI Camera's Sony IMX500 image sensor while dropping the solar panel entirely.

The latest version effectively combines the two, using the same Raspberry Pi Zero 2 W and Raspberry Pi AI Camera but with a return to solar charging rather than relying on manual intervention. "[The last version] could operate through daylight hours before being taken back inside to charge," Ditria notes. "Having to collect it and charge it and remember to put it outside every night was annoying." A large solar panel handled the energy harvesting, but off-the-shelf solutions for getting that into a battery and the Raspberry Pi proved inadequate — leading to Ditria and a friend to design their own Hardware Attached on Top (HAT) add-on, the PV Pi.

"It has true maximum power-point tracking [and] battery charging up to a peak of 10A for 12V lithium-iron-phosophate [LiFePO₄] battery packs, which means this caps out at about 140W of charging with sufficient cooling," Ditria says of the PV Pi HAT. "As for the solar input, the PV Pi works with 12V and most 24V solar panels, so it's quite flexible. And, importantly for remote Raspberry Pi projects, it can communicate directly with the Raspberry Pi to report things like battery voltage and charge current, and also has a real-time clock for power scheduling and more."

For the neural network model, Ditria trained it for 52 local species to minimize false positives. "I collected and labeled a bunch of images from online sources," he explains, "and then trained and exported a YOLO-11N [object detection] model. I've also included that bird model in my GitHub repo, that's exported and ready to run on the [Sony] IMX500 camera [module]."

Project source code is available on GitHub under an unspecified open-source license; 3D print files for the enclosure are available on Maker World under the site's Standard Digital File License.